用P-BP预测网络模型预测通信网络指标

结合某通信企业业务数据的特点，为其通信网络数据预测业务建立了一套通用的P BP预测网络模型。它以时间序列分析为建模依据和指导，并改变BP神经网络的学习方法，提出BP L网络用作模型中挖掘数据依赖性的工具，它的预测精度、运算速度、泛化能力明显高于BP网络。此外，P BP模型能依据历史数据自动计算最合适的预测阶数；根据业务数据特点设计的消除非平稳因素的方法，使其在平稳化的同时能很好地提高并行运算性能；用区间估计过滤异常数据，具有较强的抗干扰能力，能适应实际的工作环境。用业务数据测试该模型，得到了快速的、非常精确的预测效果和完备的预测值置信区间。

Data prediction in communication network using P-BP predicting network

A data predicting model in communication network field called P-BP was proposed. It was guided by time series analysis theory. BP-L neural network modified from BP network was proposed to improve its generalization ability, precision and speed in data predicting applications. BP-L was used as machine learning method in P-BP model. P-BP model could automatically leach abnormal data and get a most proper exponent. This model was tested for data predicting in CNET field and got a good result.